lustre/osd-ldiskfs/osd_oi.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_oi.c
  33  *
  34  * Object Index.
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_OSD
  40
  41 #include <linux/module.h>
  42
  43 /*
  44  * struct OBD_{ALLOC,FREE}*()
  45  * OBD_FAIL_CHECK
  46  */
  47 #include <obd.h>
  48 #include <obd_support.h>
  49
  50 /* fid_cpu_to_be() */
  51 #include <lustre_fid.h>
  52 #include <dt_object.h>
  53 #include <lustre_scrub.h>
  54
  55 #include "osd_oi.h"
  56 /* osd_lookup(), struct osd_thread_info */
  57 #include "osd_internal.h"
  58
  59 unsigned int osd_oi_count = OSD_OI_FID_NR;
  60 module_param(osd_oi_count, int, 0444);
  61 MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
  62
  63 static struct dt_index_features oi_feat = {
  64         .dif_flags       = DT_IND_UPDATE,
  65         .dif_recsize_min = sizeof(struct osd_inode_id),
  66         .dif_recsize_max = sizeof(struct osd_inode_id),
  67         .dif_ptrsize     = 4
  68 };
  69
  70 #define OSD_OI_NAME_BASE        "oi.16"
  71
  72 static void osd_oi_table_put(struct osd_thread_info *info,
  73                              struct osd_oi **oi_table, unsigned oi_count)
  74 {
  75         struct iam_container *bag;
  76         int                   i;
  77
  78         for (i = 0; i < oi_count; i++) {
  79                 if (oi_table[i] == NULL)
  80                         continue;
  81
  82                 LASSERT(oi_table[i]->oi_inode != NULL);
  83
  84                 bag = &(oi_table[i]->oi_dir.od_container);
  85                 if (bag->ic_object == oi_table[i]->oi_inode)
  86                         iam_container_fini(bag);
  87                 iput(oi_table[i]->oi_inode);
  88                 oi_table[i]->oi_inode = NULL;
  89                 OBD_FREE_PTR(oi_table[i]);
  90                 oi_table[i] = NULL;
  91         }
  92 }
  93
  94 static int osd_oi_index_create_one(struct osd_thread_info *info,
  95                                    struct osd_device *osd, const char *name,
  96                                    struct dt_index_features *feat)
  97 {
  98         const struct lu_env             *env = info->oti_env;
  99         struct osd_inode_id             *id  = &info->oti_id;
 100         struct buffer_head              *bh;
 101         struct inode                    *inode;
 102         struct ldiskfs_dir_entry_2      *de;
 103         struct dentry                   *dentry;
 104         struct super_block              *sb  = osd_sb(osd);
 105         struct inode                    *dir = sb->s_root->d_inode;
 106         handle_t                        *jh;
 107         int                              rc;
 108
 109         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 110         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
 111         if (!IS_ERR(bh)) {
 112                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 113                 brelse(bh);
 114                 inode = osd_iget(info, osd, id);
 115                 if (!IS_ERR(inode)) {
 116                         iput(inode);
 117                         inode = ERR_PTR(-EEXIST);
 118                 }
 119                 return PTR_ERR(inode);
 120         }
 121
 122         if (osd->od_dt_dev.dd_rdonly)
 123                 RETURN(-EROFS);
 124
 125         jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
 126         if (IS_ERR(jh))
 127                 return PTR_ERR(jh);
 128
 129         inode = ldiskfs_create_inode(jh, dir, (S_IFREG | S_IRUGO | S_IWUSR),
 130                                      NULL);
 131         if (IS_ERR(inode)) {
 132                 ldiskfs_journal_stop(jh);
 133                 return PTR_ERR(inode);
 134         }
 135
 136         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 137         unlock_new_inode(inode);
 138
 139         if (feat->dif_flags & DT_IND_VARKEY)
 140                 rc = iam_lvar_create(inode, feat->dif_keysize_max,
 141                                      feat->dif_ptrsize, feat->dif_recsize_max,
 142                                      jh);
 143         else
 144                 rc = iam_lfix_create(inode, feat->dif_keysize_max,
 145                                      feat->dif_ptrsize, feat->dif_recsize_max,
 146                                      jh);
 147         dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
 148         rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
 149         ldiskfs_journal_stop(jh);
 150         iput(inode);
 151         return rc;
 152 }
 153
 154 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
 155                                        struct osd_device *osd,
 156                                        const char *name,
 157                                        struct dt_index_features *f,
 158                                        bool create)
 159 {
 160         struct dentry *dentry;
 161         struct inode  *inode;
 162         int            rc;
 163
 164         dentry = osd_lookup_one_len_unlocked(osd, name, osd_sb(osd)->s_root,
 165                                              strlen(name));
 166         if (IS_ERR(dentry))
 167                 return ERR_CAST(dentry);
 168
 169         if (dentry->d_inode) {
 170                 LASSERT(!is_bad_inode(dentry->d_inode));
 171                 inode = dentry->d_inode;
 172                 atomic_inc(&inode->i_count);
 173                 dput(dentry);
 174                 return inode;
 175         }
 176
 177         /* create */
 178         dput(dentry);
 179         shrink_dcache_parent(osd_sb(osd)->s_root);
 180         if (!create)
 181                 return ERR_PTR(-ENOENT);
 182
 183         rc = osd_oi_index_create_one(info, osd, name, f);
 184         if (rc)
 185                 return ERR_PTR(rc);
 186
 187         dentry = osd_lookup_one_len_unlocked(osd, name, osd_sb(osd)->s_root,
 188                                              strlen(name));
 189         if (IS_ERR(dentry))
 190                 return ERR_CAST(dentry);
 191
 192         if (dentry->d_inode) {
 193                 LASSERT(!is_bad_inode(dentry->d_inode));
 194                 inode = dentry->d_inode;
 195                 atomic_inc(&inode->i_count);
 196                 dput(dentry);
 197                 return inode;
 198         }
 199
 200         return ERR_PTR(-ENOENT);
 201 }
 202
 203 /**
 204  * Open an OI(Ojbect Index) container.
 205  *
 206  * \param       name    Name of OI container
 207  * \param       objp    Pointer of returned OI
 208  *
 209  * \retval      0       success
 210  * \retval      -ve     failure
 211  */
 212 static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
 213                        char *name, struct osd_oi **oi_slot, bool create)
 214 {
 215         struct osd_directory *dir;
 216         struct iam_container *bag;
 217         struct inode         *inode;
 218         struct osd_oi        *oi;
 219         int                   rc;
 220
 221         ENTRY;
 222
 223         oi_feat.dif_keysize_min = sizeof(struct lu_fid);
 224         oi_feat.dif_keysize_max = sizeof(struct lu_fid);
 225
 226         inode = osd_oi_index_open(info, osd, name, &oi_feat, create);
 227         if (IS_ERR(inode))
 228                 RETURN(PTR_ERR(inode));
 229
 230         if (!osd->od_dt_dev.dd_rdonly) {
 231                 /* 'What the @fid is' is not imporatant, because these objects
 232                  * have no OI mappings, and only are visible inside the OSD.*/
 233                 lu_igif_build(&info->oti_fid, inode->i_ino,
 234                               inode->i_generation);
 235                 rc = osd_ea_fid_set(info, inode, &info->oti_fid,
 236                                     LMAC_NOT_IN_OI, 0);
 237                 if (rc)
 238                         GOTO(out_inode, rc);
 239         }
 240
 241         OBD_ALLOC_PTR(oi);
 242         if (oi == NULL)
 243                 GOTO(out_inode, rc = -ENOMEM);
 244
 245         oi->oi_inode = inode;
 246         dir = &oi->oi_dir;
 247
 248         bag = &dir->od_container;
 249         rc = iam_container_init(bag, &dir->od_descr, inode);
 250         if (rc < 0)
 251                 GOTO(out_free, rc);
 252
 253         rc = iam_container_setup(bag);
 254         if (rc < 0)
 255                 GOTO(out_container, rc);
 256
 257         *oi_slot = oi;
 258         RETURN(0);
 259
 260 out_container:
 261         iam_container_fini(bag);
 262 out_free:
 263         OBD_FREE_PTR(oi);
 264 out_inode:
 265         iput(inode);
 266         return rc;
 267 }
 268
 269 /**
 270  * Open OI(Object Index) table.
 271  * If \a oi_count is zero, which means caller doesn't know how many OIs there
 272  * will be, this function can either return 0 for new filesystem, or number
 273  * of OIs on existed filesystem.
 274  *
 275  * If \a oi_count is non-zero, which means caller does know number of OIs on
 276  * filesystem, this function should return the exactly same number on
 277  * success, or error code in failure.
 278  *
 279  * \param     oi_count  Number of expected OI containers
 280  * \param     create    Create OIs if doesn't exist
 281  *
 282  * \retval    +ve       number of opened OI containers
 283  * \retval      0       no OI containers found
 284  * \retval    -ve       failure
 285  */
 286 static int
 287 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
 288                   struct osd_oi **oi_table, unsigned oi_count, bool create)
 289 {
 290         struct scrub_file *sf = &osd->od_scrub.os_scrub.os_file;
 291         int count = 0;
 292         int rc = 0;
 293         int i;
 294         ENTRY;
 295
 296         /* NB: oi_count != 0 means that we have already created/known all OIs
 297          * and have known exact number of OIs. */
 298         LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
 299
 300         for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
 301                 char name[sizeof(OSD_OI_NAME_BASE) + 3 * sizeof(i) + 1];
 302
 303                 if (oi_table[i] != NULL) {
 304                         count++;
 305                         continue;
 306                 }
 307
 308                 snprintf(name, sizeof(name), "%s.%d", OSD_OI_NAME_BASE, i);
 309                 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
 310                 if (rc == 0) {
 311                         count++;
 312                         continue;
 313                 }
 314
 315                 if (rc == -ENOENT && create == false) {
 316                         if (oi_count == 0)
 317                                 return count;
 318
 319                         rc = 0;
 320                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 321                         continue;
 322                 }
 323
 324                 CERROR("%s: can't open %s: rc = %d\n",
 325                        osd_dev2name(osd), name, rc);
 326                 if (oi_count > 0)
 327                         CERROR("%s: expect to open total %d OI files.\n",
 328                                osd_dev2name(osd), oi_count);
 329                 break;
 330         }
 331
 332         if (rc < 0) {
 333                 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
 334                 count = rc;
 335         }
 336
 337         RETURN(count);
 338 }
 339
 340 static int osd_remove_oi_one(struct osd_device *osd, struct dentry *parent,
 341                              const char *name, int namelen)
 342 {
 343         struct dentry *child;
 344         int rc;
 345
 346         child = osd_lookup_one_len_unlocked(osd, name, parent, namelen);
 347         if (IS_ERR(child)) {
 348                 rc = PTR_ERR(child);
 349         } else {
 350                 rc = ll_vfs_unlink(parent->d_inode, child);
 351                 dput(child);
 352         }
 353
 354         return rc == -ENOENT ? 0 : rc;
 355 }
 356
 357 static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
 358 {
 359         char name[16];
 360         int namelen;
 361         int rc;
 362         int i;
 363
 364         if (osd->od_dt_dev.dd_rdonly)
 365                 RETURN(-EROFS);
 366
 367         for (i = 0; i < OSD_OI_FID_NR_MAX; i++) {
 368                 namelen = snprintf(name, sizeof(name), "%s.%d",
 369                                    OSD_OI_NAME_BASE, i);
 370                 rc = osd_remove_oi_one(osd, osd_sb(osd)->s_root, name, namelen);
 371                 if (rc != 0) {
 372                         CERROR("%s: fail to remove the stale OI file %s: "
 373                                "rc = %d\n", osd_dev2name(osd), name, rc);
 374                         return rc;
 375                 }
 376         }
 377
 378         namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
 379         rc = osd_remove_oi_one(osd, osd_sb(osd)->s_root, name, namelen);
 380         if (rc != 0)
 381                 CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
 382                        osd_dev2name(osd), name, rc);
 383
 384         return rc;
 385 }
 386
 387 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
 388                 bool restored)
 389 {
 390         struct lustre_scrub *scrub = &osd->od_scrub.os_scrub;
 391         struct scrub_file *sf = &scrub->os_file;
 392         struct osd_oi **oi;
 393         int count;
 394         int rc;
 395         ENTRY;
 396
 397         if (unlikely(sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
 398                 LCONSOLE_WARN("%s: Invalid OI count in scrub file %d\n",
 399                               osd_dev2name(osd), sf->sf_oi_count);
 400                 sf->sf_oi_count = 0;
 401         }
 402
 403         if (restored) {
 404                 rc = osd_remove_ois(info, osd);
 405                 if (rc)
 406                         RETURN(rc);
 407         }
 408
 409         OBD_ALLOC_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 410         if (oi == NULL)
 411                 RETURN(-ENOMEM);
 412
 413         /* try to open existing multiple OIs first */
 414         count = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
 415         if (count < 0)
 416                 GOTO(out, rc = count);
 417
 418         if (count > 0) {
 419                 if (count == sf->sf_oi_count)
 420                         GOTO(out, rc = count);
 421
 422                 if (sf->sf_oi_count == 0) {
 423                         if (likely((count & (count - 1)) == 0))
 424                                 GOTO(out, rc = count);
 425
 426                         LCONSOLE_WARN("%s: invalid oi count %d, remove them, "
 427                                       "then set it to %d\n", osd_dev2name(osd),
 428                                       count, osd_oi_count);
 429                         osd_oi_table_put(info, oi, count);
 430                         rc = osd_remove_ois(info, osd);
 431                         if (rc)
 432                                 GOTO(out, rc);
 433
 434                         sf->sf_oi_count = osd_oi_count;
 435                 }
 436
 437                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 438                 count = sf->sf_oi_count;
 439                 goto create;
 440         }
 441
 442         /* if previous failed then try found single OI from old filesystem */
 443         rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
 444         if (rc == 0) { /* found single OI from old filesystem */
 445                 count = 1;
 446                 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
 447                 if (sf->sf_success_count == 0)
 448                         /* XXX: There is one corner case that if the OI_scrub
 449                          *      file crashed or lost and we regard it upgrade,
 450                          *      then we allow IGIF lookup to bypass OI files.
 451                          *
 452                          *      The risk is that osd_fid_lookup() may found
 453                          *      a wrong inode with the given IGIF especially
 454                          *      when the MDT has performed file-level backup
 455                          *      and restored after former upgrading from 1.8
 456                          *      to 2.x. Fortunately, the osd_fid_lookup()can
 457                          *      verify the inode to decrease the risk. */
 458                         scrub_file_reset(scrub, osd->od_uuid, SF_UPGRADE);
 459                 GOTO(out, rc = 1);
 460         } else if (rc != -ENOENT) {
 461                 CERROR("%s: can't open %s: rc = %d\n",
 462                        osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
 463                 GOTO(out, rc);
 464         }
 465
 466         if (sf->sf_oi_count > 0) {
 467                 int i;
 468
 469                 count = sf->sf_oi_count;
 470                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 471                 for (i = 0; i < count; i++)
 472                         ldiskfs_set_bit(i, sf->sf_oi_bitmap);
 473                 scrub_file_reset(scrub, osd->od_uuid, SF_RECREATED);
 474         } else {
 475                 count = sf->sf_oi_count = osd_oi_count;
 476         }
 477
 478 create:
 479         rc = scrub_file_store(info->oti_env, scrub);
 480         if (rc < 0) {
 481                 osd_oi_table_put(info, oi, count);
 482                 GOTO(out, rc);
 483         }
 484
 485         /* No OIs exist, new filesystem, create OI objects */
 486         rc = osd_oi_table_open(info, osd, oi, count, true);
 487         LASSERT(ergo(rc >= 0, rc == count));
 488
 489         GOTO(out, rc);
 490
 491 out:
 492         if (rc < 0) {
 493                 OBD_FREE_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 494         } else {
 495                 LASSERTF((rc & (rc - 1)) == 0, "Invalid OI count %d\n", rc);
 496
 497                 osd->od_oi_table = oi;
 498                 osd->od_oi_count = rc;
 499                 if (sf->sf_oi_count != rc) {
 500                         sf->sf_oi_count = rc;
 501                         rc = scrub_file_store(info->oti_env, scrub);
 502                         if (rc < 0) {
 503                                 osd_oi_table_put(info, oi, count);
 504                                 OBD_FREE_PTR_ARRAY(oi, OSD_OI_FID_NR_MAX);
 505                         }
 506                 } else {
 507                         rc = 0;
 508                 }
 509         }
 510
 511         return rc;
 512 }
 513
 514 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 515 {
 516         if (unlikely(!osd->od_oi_table))
 517                 return;
 518
 519         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 520
 521         OBD_FREE_PTR_ARRAY(osd->od_oi_table, OSD_OI_FID_NR_MAX);
 522         osd->od_oi_table = NULL;
 523 }
 524
 525 static inline int fid_is_fs_root(const struct lu_fid *fid)
 526 {
 527         /* Map root inode to special local object FID */
 528         return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 529                          fid_oid(fid) == OSD_FS_ROOT_OID));
 530 }
 531
 532 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 533                              struct osd_oi *oi, struct dt_rec *rec,
 534                              const struct dt_key *key)
 535 {
 536         struct iam_container  *bag;
 537         struct iam_iterator   *it = &oti->oti_idx_it;
 538         struct iam_path_descr *ipd;
 539         int                    rc;
 540         ENTRY;
 541
 542         LASSERT(oi);
 543         LASSERT(oi->oi_inode);
 544
 545         bag = &oi->oi_dir.od_container;
 546         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 547         if (IS_ERR(ipd))
 548                 RETURN(-ENOMEM);
 549
 550         /* got ipd now we can start iterator. */
 551         iam_it_init(it, bag, 0, ipd);
 552
 553         rc = iam_it_get(it, (struct iam_key *)key);
 554         if (rc > 0)
 555                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
 556         iam_it_put(it);
 557         iam_it_fini(it);
 558         osd_ipd_put(oti->oti_env, bag, ipd);
 559
 560         LINVRNT(osd_invariant(obj));
 561
 562         RETURN(rc);
 563 }
 564
 565 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
 566                   const struct lu_fid *fid, enum oi_check_flags flags)
 567 {
 568         struct lu_seq_range     *range = &info->oti_seq_range;
 569         int                     rc;
 570         ENTRY;
 571
 572         if (flags & OI_KNOWN_ON_OST)
 573                 RETURN(1);
 574
 575         if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
 576                      fid_is_llog(fid)) || fid_is_name_llog(fid) ||
 577                      fid_is_quota(fid))
 578                 RETURN(0);
 579
 580         if (fid_is_idif(fid) || fid_is_last_id(fid))
 581                 RETURN(1);
 582
 583         if (!(flags & OI_CHECK_FLD))
 584                 RETURN(0);
 585
 586         if (osd_seq_site(osd)->ss_server_fld == NULL)
 587                 RETURN(0);
 588
 589         rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
 590         if (rc != 0) {
 591                 /* During upgrade, OST FLDB might not be loaded because
 592                  * OST FLDB is not created until 2.6, so if some DNE
 593                  * filesystem upgrade from 2.5 to 2.7/2.8, they will
 594                  * not be able to find the sequence from local FLDB
 595                  * cache see fld_index_init(). */
 596                 if (rc == -ENOENT && osd->od_is_ost)
 597                         RETURN(1);
 598
 599                 if (rc != -ENOENT)
 600                         CERROR("%s: lookup FLD "DFID": rc = %d\n",
 601                                osd_name(osd), PFID(fid), rc);
 602                 RETURN(0);
 603         }
 604
 605         if (fld_range_is_ost(range))
 606                 RETURN(1);
 607
 608         RETURN(0);
 609 }
 610
 611 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 612                            const struct lu_fid *fid, struct osd_inode_id *id)
 613 {
 614         struct lu_fid *oi_fid = &info->oti_fid2;
 615         int            rc;
 616
 617         fid_cpu_to_be(oi_fid, fid);
 618         rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
 619                                (const struct dt_key *)oi_fid);
 620         if (rc > 0) {
 621                 osd_id_unpack(id, id);
 622                 rc = 0;
 623         } else if (rc == 0) {
 624                 rc = -ENOENT;
 625         }
 626         return rc;
 627 }
 628
 629 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 630                   const struct lu_fid *fid, struct osd_inode_id *id,
 631                   enum oi_check_flags flags)
 632 {
 633         if (unlikely(fid_is_last_id(fid)))
 634                 return osd_obj_spec_lookup(info, osd, fid, id, flags);
 635
 636         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 637                 return osd_obj_map_lookup(info, osd, fid, id);
 638
 639
 640         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 641                 int rc;
 642                 if (fid_is_fs_root(fid)) {
 643                         osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
 644                                    osd_sb(osd)->s_root->d_inode->i_generation);
 645                         return 0;
 646                 }
 647                 if (unlikely(fid_is_acct(fid)))
 648                         return osd_acct_obj_lookup(info, osd, fid, id);
 649
 650                 /* For other special FIDs, try OI first, then do spec lookup */
 651                 rc = __osd_oi_lookup(info, osd, fid, id);
 652                 if (rc == -ENOENT)
 653                         return osd_obj_spec_lookup(info, osd, fid, id, flags);
 654                 return rc;
 655         }
 656
 657         if (!osd->od_igif_inoi && fid_is_igif(fid)) {
 658                 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
 659                 return 0;
 660         }
 661
 662         return __osd_oi_lookup(info, osd, fid, id);
 663 }
 664
 665 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 666                              const struct dt_rec *rec, const struct dt_key *key,
 667                              handle_t *th, bool insert)
 668 {
 669         struct iam_container    *bag;
 670         struct iam_path_descr   *ipd;
 671         int                     rc;
 672         ENTRY;
 673
 674         LASSERT(oi);
 675         LASSERT(oi->oi_inode);
 676         dquot_initialize(oi->oi_inode);
 677
 678         bag = &oi->oi_dir.od_container;
 679         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 680         if (unlikely(ipd == NULL))
 681                 RETURN(-ENOMEM);
 682
 683         LASSERT(th != NULL);
 684         LASSERT(th->h_transaction != NULL);
 685         if (insert)
 686                 rc = iam_insert(th, bag, (const struct iam_key *)key,
 687                                 (const struct iam_rec *)rec, ipd);
 688         else
 689                 rc = iam_update(th, bag, (const struct iam_key *)key,
 690                                 (const struct iam_rec *)rec, ipd);
 691         osd_ipd_put(oti->oti_env, bag, ipd);
 692         LINVRNT(osd_invariant(obj));
 693         RETURN(rc);
 694 }
 695
 696 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 697                   const struct lu_fid *fid, const struct osd_inode_id *id,
 698                   handle_t *th, enum oi_check_flags flags, bool *exist)
 699 {
 700         struct lu_fid       *oi_fid = &info->oti_fid2;
 701         struct osd_inode_id *oi_id  = &info->oti_id2;
 702         int                  rc     = 0;
 703
 704         if (unlikely(fid_is_last_id(fid)))
 705                 return osd_obj_spec_insert(info, osd, fid, id, th);
 706
 707         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 708                 return osd_obj_map_insert(info, osd, fid, id, th);
 709
 710         fid_cpu_to_be(oi_fid, fid);
 711         osd_id_pack(oi_id, id);
 712         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 713                                (const struct dt_rec *)oi_id,
 714                                (const struct dt_key *)oi_fid, th, true);
 715         if (rc != 0) {
 716                 struct inode *inode;
 717                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 718
 719                 if (rc != -EEXIST)
 720                         return rc;
 721
 722                 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
 723                 if (rc != 0)
 724                         return rc;
 725
 726                 if (unlikely(osd_id_eq(id, oi_id)))
 727                         return 1;
 728
 729                 /* Check whether the mapping for oi_id is valid or not. */
 730                 inode = osd_iget(info, osd, oi_id);
 731                 if (IS_ERR(inode)) {
 732                         rc = PTR_ERR(inode);
 733                         if (rc == -ENOENT || rc == -ESTALE)
 734                                 goto update;
 735                         return rc;
 736                 }
 737
 738                 /* The EA inode should NOT be in OI, old OI scrub may added
 739                  * such OI mapping by wrong, replace it. */
 740                 if (unlikely(osd_is_ea_inode(inode))) {
 741                         iput(inode);
 742                         goto update;
 743                 }
 744
 745                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 746                                  &info->oti_ost_attrs);
 747                 iput(inode);
 748                 if (rc == -ENODATA)
 749                         goto update;
 750
 751                 if (rc != 0)
 752                         return rc;
 753
 754                 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 755                     lu_fid_eq(fid, &lma->lma_self_fid)) {
 756                         CERROR("%s: the FID "DFID" is used by two objects: "
 757                                "%u/%u %u/%u\n", osd_dev2name(osd),
 758                                PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
 759                                id->oii_ino, id->oii_gen);
 760                         return -EEXIST;
 761                 }
 762
 763 update:
 764                 osd_id_pack(oi_id, id);
 765                 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 766                                         (const struct dt_rec *)oi_id,
 767                                         (const struct dt_key *)oi_fid, th, false);
 768                 if (rc != 0)
 769                         return rc;
 770
 771                 if (exist != NULL)
 772                         *exist = true;
 773         }
 774
 775         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 776                 rc = osd_obj_spec_insert(info, osd, fid, id, th);
 777         return rc;
 778 }
 779
 780 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
 781                              const struct dt_key *key, handle_t *th)
 782 {
 783         struct iam_container    *bag;
 784         struct iam_path_descr   *ipd;
 785         int                      rc;
 786         ENTRY;
 787
 788         LASSERT(oi);
 789         LASSERT(oi->oi_inode);
 790         dquot_initialize(oi->oi_inode);
 791
 792         bag = &oi->oi_dir.od_container;
 793         ipd = osd_idx_ipd_get(oti->oti_env, bag);
 794         if (unlikely(ipd == NULL))
 795                 RETURN(-ENOMEM);
 796
 797         LASSERT(th != NULL);
 798         LASSERT(th->h_transaction != NULL);
 799
 800         rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
 801         osd_ipd_put(oti->oti_env, bag, ipd);
 802         LINVRNT(osd_invariant(obj));
 803         RETURN(rc);
 804 }
 805
 806 int osd_oi_delete(struct osd_thread_info *info,
 807                   struct osd_device *osd, const struct lu_fid *fid,
 808                   handle_t *th, enum oi_check_flags flags)
 809 {
 810         struct lu_fid *oi_fid = &info->oti_fid2;
 811
 812         /* clear idmap cache */
 813         if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
 814                 fid_zero(&info->oti_cache.oic_fid);
 815
 816         if (fid_is_last_id(fid))
 817                 return 0;
 818
 819         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 820                 return osd_obj_map_delete(info, osd, fid, th);
 821
 822         fid_cpu_to_be(oi_fid, fid);
 823         return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
 824                                  (const struct dt_key *)oi_fid, th);
 825 }
 826
 827 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
 828                   const struct lu_fid *fid, const struct osd_inode_id *id,
 829                   handle_t *th, enum oi_check_flags flags)
 830 {
 831         struct lu_fid       *oi_fid = &info->oti_fid2;
 832         struct osd_inode_id *oi_id  = &info->oti_id2;
 833         int                  rc     = 0;
 834
 835         if (unlikely(fid_is_last_id(fid)))
 836                 return osd_obj_spec_update(info, osd, fid, id, th);
 837
 838         if (fid_is_llog(fid) || fid_is_on_ost(info, osd, fid, flags))
 839                 return osd_obj_map_update(info, osd, fid, id, th);
 840
 841         fid_cpu_to_be(oi_fid, fid);
 842         osd_id_pack(oi_id, id);
 843         rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
 844                                (const struct dt_rec *)oi_id,
 845                                (const struct dt_key *)oi_fid, th, false);
 846         if (rc != 0)
 847                 return rc;
 848
 849         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
 850                 rc = osd_obj_spec_update(info, osd, fid, id, th);
 851         return rc;
 852 }
 853
 854 int osd_oi_mod_init(void)
 855 {
 856         if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
 857                 osd_oi_count = OSD_OI_FID_NR;
 858
 859         if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
 860                 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
 861                               osd_oi_count, size_roundup_power2(osd_oi_count));
 862                 osd_oi_count = size_roundup_power2(osd_oi_count);
 863         }
 864
 865         return 0;
 866 }